Add sort benchmark #278
Closed
Add sort benchmark #278
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
aprokop
added
testing
aprokop
force-pushed
the
sort_benchmark
branch
2 times, most recently
from
441f39f
to
cca7303
Compare
dalg24
reviewed
Apr 21, 2020
|
So far, I cannot reproduce results from here, at least on my workstation. |
|
Do you see anything similar to my observations in #222? |
|
I think yes, for smaller sizes Serial outperforms any Cuda, for both sort and permutations. Though that was on random data. |
dalg24
reviewed
Apr 24, 2020
This is a modified version of the original PSS (https://github.com/wjakob/pss) taken from https://github.com/SNLComputation/omega_h/tree/144a1c16b5f4c8dd66e55e8d43e9c37af0b5890b/tpl/pss
|
Looks like there is some guard missing: |
benchmarks/sort/sort_benchmark.cpp
Outdated
|
|
||
| Kokkos::BinSort<ViewType, CompType, typename ViewType::device_type, | ||
| SizeType> | ||
| bin_sort(view, CompType(n / 2, result.min_val, result.max_val), true); |
There was a problem hiding this comment.
Can you remind me why we set the maximum number of bins to n/2 here?
There was a problem hiding this comment.
I don't recall, maybe @dalg24 remembers. Maybe this is one of the things we need to look at.
benchmarks/sort/sort_benchmark.cpp
Outdated
| int const n = in.extent(0); | ||
| Kokkos::parallel_for( | ||
| Kokkos::RangePolicy<ExecutionSpace>(0, n), | ||
| KOKKOS_LAMBDA(int const i) { out(permute(i)) = in(i); }); |
There was a problem hiding this comment.
It might also be interesting to see if the read access is non-consecutive and write is.
There was a problem hiding this comment.
Yes, I was thinking the same thing, maybe introduce reverse permutation option.
|
I have a new branch |
So you obtained some timings already? Did you save your numbers somewhere? |
|
That's some of the numbers I am seeing |
This avoids doing extra work (e.g., creating extra views) for regular use case (memory is accessible from execution space) which affects timings for small sizes.
benchmarks/sort/sort_benchmark.cpp
Outdated
| int const n = view.extent(0); | ||
|
|
||
| auto view_mirror = | ||
| Kokkos::create_mirror_view_and_copy(execution_space{}, view); |
There was a problem hiding this comment.
If we use instances, it would make sense to at least reuse it if possible.
There was a problem hiding this comment.
Not sure what you mean.
dalg24
reviewed
May 1, 2020
| int const n = view.extent(0); | ||
|
|
||
| auto begin_ptr = thrust::device_ptr<ValueType>(view.data()); | ||
| auto end_ptr = thrust::device_ptr<ValueType>(view.data() + n); |
There was a problem hiding this comment.
thrust::device_ptr<value_type> begin(Kokkos::View<value_type *, memory_space> const &v) { return {v.data()}; }
thrust::device_ptr<value_type> end(Kokkos::View<value_type *, memory_space> const &v) { return {v.data() + v.size()}; }
|
Some results on So StdSort is faster up to 1500 values. Except for really small values PSS_OpenMP is the fastest OpenMP variant up to around 32000 values after that |
|
Results from the previous comment for easier reading (just Time column):
|
|
Results on I am not quite sure what happened for 4000, 8000 and 16000. Possibly some other process running despite requesting one full node. |
|
These are the Basically showing for |
|
Closing for now due to the lack of activity and interest. Will resurrect if need arises. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
No description provided.